Ask Your Question
0

image pattern recognition

asked 2013-03-23 07:38:05 -0500

Victor1234 gravatar image

What do you think, what alghoritms developers used in this video http://vimeo.com/50747223, why it so perfect?

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
2

answered 2013-03-23 08:18:53 -0500

Guanta gravatar image

updated 2013-03-23 08:52:09 -0500

The text below the video says: "By overlaying the camera of a mobile device onto objects, the ARART-system will detect images and objects that are registered and will display an animated image on top of said object." .

So, I guess it goes in two steps, first it tries to find the image in an image database if it finds it, it displays a pre-defined random (probably there exist more than just one) video-sequence aligned (i.e. with the correct perspective) to the camera. If it doesn't find the image, then it tries to identify objects in the image, takes one and displays (probably) a random sequence which is connected to that object. If nothing works maybe some color distortion or other effects are applied.

So the most difficult part is the detection of the image/objeect: Maybe they used a global image descriptor, see http://answers.opencv.org/question/9271/global-image-feature-implementation/ and http://answers.opencv.org/question/8677/image-comparison-with-a-database#8686.

For object detection they could use basically the same approach by tiling the image in overlapping windows. Finally, the alignment with the camera can be achieved by keypoint/descriptor matching. Then this distortian is applied to the predefined video-sequence/ animation.

edit flag offensive delete link more

Comments

@Guanta, i try to use SURF, but it gave too many false point matches, it is possible to get roi mask by some other alghorithm?

Victor1234 gravatar imageVictor1234 ( 2013-03-25 15:18:54 -0500 )edit

What are you trying to accomplish? In which sense false positives? Maybe you want to ask a separate question for that.

Guanta gravatar imageGuanta ( 2013-03-25 15:33:07 -0500 )edit

I get some results with SURF and FLANN, but i want to try some machine learning like k-nearest instead of them, where i can find good tutorial to train and use k-nearest algorithm for planar image recognision?

Victor1234 gravatar imageVictor1234 ( 2013-03-30 04:39:22 -0500 )edit

What do you mean by 'k-nearest'? I know only 'K-nearest neighbor', which means that you are searching for the k-nearest neighbor feature vector (in feature space) which you can approximate efficiently with FLANN. To answer your question: Maybe this good open source computer vision book helps you further: http://szeliski.org/Book/ . It gives good ideas and points to important literature.

Guanta gravatar imageGuanta ( 2013-03-30 07:04:40 -0500 )edit
Login/Signup to Answer

Question Tools

Stats

Asked: 2013-03-23 07:38:05 -0500

Seen: 915 times

Last updated: Mar 23 '13